This project provides a browser automation playground powered by Google's Gemini models and Browserbase infrastructure. It targets engineers and power users seeking to automate web tasks using natural language, offering a benefit of intuitive, AI-driven control over browser interactions with real-time feedback.

How It Works

The core approach leverages Gemini's advanced Computer Use Agent (CUA) capabilities, specifically Gemini 2.5 CUA or Gemini 3 Flash, to interpret natural language commands. These commands are then executed via the Stagehand agent framework, which interfaces with Browserbase's real browser infrastructure. This combination allows for complex web navigation and interaction driven purely by descriptive text, with progress streamed live using Server-Sent Events (SSE).

Quick Start & Requirements

• Primary install: Clone the repository, cd into it, and run pnpm install.
• Prerequisites: Node.js 18.x or later, pnpm 10.x or later (recommended). Requires API keys for Google AI Studio (GOOGLE_API_KEY) and Browserbase (BROWSERBASE_API_KEY, BROWSERBASE_PROJECT_ID). Optional keys for PostHog analytics and Vercel Edge Config.
• Run: Execute pnpm dev to start the development server.
• Links: Repository available at https://github.com/browserbase/gemini-browser.

Highlighted Details

• Supports switching between Gemini 2.5 Computer Use (CUA) and Gemini 3 Flash models.
• Enables natural language commands for executing browser tasks.
• Features real-time streaming of agent feedback and progress via SSE.
• Provides persistent browser sessions with automatic viewport management.

Maintenance & Community

This is described as a demo project. Acknowledgments include Browserbase, Stagehand, Google AI Studio, and Vercel. No specific community channels (like Discord/Slack) or roadmap details are provided in the README.

Licensing & Compatibility

The project is licensed under the MIT license. This permissive license generally allows for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

As a demo project, its primary purpose is showcasing Gemini CUA capabilities. It requires obtaining and configuring multiple third-party API keys, which may involve associated costs or sign-up processes. The reliance on Browserbase infrastructure implies potential usage limits or fees. The use of preview AI models suggests that functionality may evolve or change.